Same context as issue #182: benchmarking v0.4.0 (`a52f990`) directly, findings validated by an agent (as I'm not fluent enough in ruts, yet), all offline.

`HaltReason` has a single variant ([types.rs#L239-L244](https://github.com/denoland/celld/blob/a52f9905425bc41134d817694bdc2c50bcc5e856/crates/logic/types.rs#L239-L244)) and `fence_node` emits it on every path ([lib.rs#L2810-L2817](https://github.com/denoland/celld/blob/a52f9905425bc41134d817694bdc2c50bcc5e856/crates/logic/lib.rs#L2810-L2817)), so the immediate fence on a missing/different read-back ([lib.rs#L2087-L2092](https://github.com/denoland/celld/blob/a52f9905425bc41134d817694bdc2c50bcc5e856/crates/logic/lib.rs#L2087-L2092)) logs the same line as the watchdog ([actor.rs#L3734-L3740](https://github.com/denoland/celld/blob/a52f9905425bc41134d817694bdc2c50bcc5e856/crates/celld/actor.rs#L3734-L3740)).

### Reproduction

just 60s TTL, delete the node's own lease object after acquisition, wait for the next renewal.

``` 11:59:04.263 acquire applied; expires_ms=1788264004255 11:59:05.457 lease object removed out of band 11:59:27.595 renewal started; prior_authority_headroom_ms=36660 11:59:27.596 renewal rejected 11:59:27.596 SELF-FENCE: node lease not renewed within TTL — halting (exit 3) ```

Fencing at once is right, the record did move. It's the reported cause that is off: 36.7s of the TTL were still there, and [`docs/guarantees.md`](https://github.com/denoland/celld/blob/a52f9905425bc41134d817694bdc2c50bcc5e856/docs/guarantees.md#L208-L224) says the line names the cause.

### Additional

Related wish while I'm here: when a node fences there's no way to tell from its own output whether the read-back came back missing, different, or as a read error. The S3 side is only visible if you put something in front of the store, which is a lot of setup for a diagnosis. Some built-in request-level logging for the store adapter, opt-in, even just the outcome class on the lease path, would make these self-diagnosable.

WDYT?